Reservoir Fine Evaluation of Horizontal Well Multi-Stage Fracturing Based on Unsupervised Cluster Machine Learning

The accuracy of comprehensive evaluation is low, the effect of multi-stage and multi-cluster design is poor, and it is difficult to ensure the maximum stimulation efficiency. The comprehensive quality of reservoirs is quantitatively evaluated by considering the physical properties and fracability of reservoirs based on logging data. This evaluation integrated reservoir quality (RQ) and completion quality (CQ) from a geology-engineering perspective, a model is established using the entropy weight method (EWM) and the analytic hierarchy process (AHP). Then, the unsupervised Gaussian Mixture Model (GMM) clustering is applied to categorize the composite quality index (CQI), resulting in the development of a novel evaluation model for multi-cluster fractured. The evaluation and clustering are conducted using Jimusaer shale oil well A in Xinjiang and Ordos Basin well B in Qingcheng as case studies. Upon comparing the results with the actual oil production, it is discovered that: the comprehensive quality can be quantitatively characterized by the evaluation model; the GMM can accurately identify good fracability zones, and the coincidence rate with the results of the oil production is as high as 83.3%, which confirms the accuracy of the proposed method, provides a good technical foundation for optimizing the location of multi-stages and multi-clusters. 1. INTRODUCTION Under the current background of unconventional oil and gas exploitation with low quality aiming at cost reduction and efficiency improvement, an important step to carry out multi-stage and cluster stimulation for low and ultralow permeability reservoirs is to determine good reservoir quality (RQ) and completion quality (CQ) (Hu et al., 2018). Usually, there are problems such as parameter diversification and inaccurate quantification in quality evaluation (Lei et al., 2019; Lei et al., 2022), and unreasonable position of multi-stage and perforation cluster, resulting in low production after reservoir stimulation (Miller et al., 2011; Wigger et al., 2014).